According to recent studies, interest rates and interest rates spreads predict economic activity (see, for example, Balke and Emery, 1994; Bernanke, 1990; Bernanke and Blinder, 1992; Friedman and Kuttner, 1992; Stock and Watson, 1989). This paper explores the characteristics of the movements in interest rates and rate spreads that are responsible for this predictive power. The paper divides interest rate and spread movements into two components: (i) ordinary time-series processes and (ii) the effects of extraordinary disturbances to those processes. A statistical, outlier search technique developed by Tsay (1988) and introduced to the economics literature by Balke and Fomby (1991) identifies the dates of extraordinary disturbances, measures their signs and magnitudes, and characterizes their persistence.
This paper focuses on the federal funds rate and two interest rate spreads, the difference between the 6-month commercial paper and 6-month T-Bill rates and the difference between the federal funds rate and the rate on 10-year Treasury bonds. For these series, positive movements correspond to contractionary monetary and financial conditions.
The predictive power of the interest rate variables is due to the extraordinary disturbances identified by the Balke-Fomby technique. These disturbances correspond to periods of monetary policy intervention. The predictive power of the interest rate series seems to derive primarily from disturbances corresponding to contractionary monetary policy.
II. IDENTIFYING EXTRAORDINARY DISTURBANCES
Following Tsay and Balke and Fomby, this paper models a time-series, [R.sub.t], as the sum of two separate processes:
(1) [R.sub.t] = [Z.sub.t] + f(t)
where [Z.sub.t] is the ordinary component of [R.sub.t] and f(t) is the effect of extraordinary disturbances to [R.sub.t]. [Z.sub.t] is modeled as an autoregressive (AR) process
(2) [Z.sub.t] = [[Theta].sub.0] + [1/[Phi](L)][a.sub.t],
where [Phi](L) is a polynomial in the lag operator, L, and [a.sub.t] is a normally distributed random variable with mean zero and variance [Mathematical Expression Omitted]. f(t) is the sum of the effects of individual disturbances.
Each extraordinary disturbance generates a time-series of effects, [f.sub.[Tau]](t), where [Tau] indexes the period in which the disturbance occurs. If a disturbance occurs at t = [Tau], then [f.sub.[Tau]]([Tau]) [not equal to] 0 and [f.sub.[Tau]](t) = 0 for t [less than] [Tau]. There are three types of dynamic effects following a disturbance: (i) a temporary one-period shock ([f.sub.[Tau]](t) = 0 except at t = [Tau]), (ii) a persistent shock ([f.sub.[Tau]](t) is non-zero for t = [Tau], [Tau] + 1, ..., but eventually goes to zero), and (iii) a permanent shock or shift in the time-series (if [f.sub.[Tau]]([Tau]) = k, then [f.sub.[Tau]]([Tau] + j) = k for all j [greater than] 0). The dynamics of the temporary and permanent cases for f(t) are clear, but there are an infinite number of possible persistent patterns. Following Balke and Fomby, assume that the persistent disturbances follow the same AR process as does [R.sub.t]. f(t) in eq. (1) is the sum of the individual [f.sub.[Tau]](t). In this study, [f.sub.[Tau]](t) = 0 for all t for most values of [Tau]. That is, extraordinary disturbances occur infrequently. Balke and Fomby describe statistical methods for identifying the periods in which extraordinary disturbances occur, for characterizing their dynamic properties, and for estimating their magnitudes. Those techniques are the basis for the division of interest rate series into ordinary and extraordinary disturbance components in this paper. (The application of the Balke and Fomby technique is described in an appendix that is available on request from the author.)
Recent research has focused on the predictive power of three types of interest rate series. These are short-term interest rates (usually the federal funds rate), a spread based on asset quality, and a spread based on term-to-maturity. …